Linear Regression.py

#================================================================================================================
#----------------------------------------------------------------------------------------------------------------
#									SIMPLE LINEAR REGRESSION
#----------------------------------------------------------------------------------------------------------------
#================================================================================================================

#Simple linear regression is applied to stock data, where the x values are time and y values are the stock closing price.
#This is not an ideal application of simple linear regression, but it suffices to be a good experiment.

import math
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import style
import pandas
import datetime

#Quandl for getting stock data
import quandl

#for plotting
plt.style.use('ggplot')

class CustomLinearRegression:
	
	def __init__(self):
		self.intercept = 0
		self.slope = 0

	#arithmetic mean
	def am(self, arr):
		tot = 0.0
		for i in arr:
			tot+= i
		return tot/len(arr)

	#finding the slope in best fit line
	def best_fit(self, dimOne, dimTwo):
		self.slope = ( (self.am(dimOne) * self.am(dimTwo) ) - self.am(dimOne*dimTwo) ) / ( self.am(dimOne)**2 - self.am(dimOne**2) ) #formula for finding slope
		return self.slope

	#finding the best fit intercept
	def y_intercept(self, dimOne ,dimTwo):
		self.intercept = self.am( dimTwo ) - ( self.slope * self.am(dimOne) )
		return self.intercept

	#predict for future values based on model
	def predict(self, ip):
		ip = np.array(ip)
		predicted = [(self.slope*param) + self.intercept for param in ip] #create a "predicted" array where the index corresponds to the index of the input
		return predicted
		
	#find the squared error
	def squared_error(self, original, model):
		return sum((model - original) **2)

	#find co-efficient of determination for R^2
	def cod(self, original, model):
		am_line = [self.am(original) for y in original]
		sq_error = self.squared_error(original, model)
		sq_error_am = self.squared_error(original, am_line)
		return 1 - (sq_error/sq_error_am) #R^2 is nothing but 1 - of squared error for our model / squared error if the model only consisted of the mean

def main():
	stk = quandl.get("WIKI/TSLA")

	simpl_linear_regression = CustomLinearRegression()

	#reset index to procure date - date was the initial default index
	stk = stk.reset_index()
	
	#Add them headers
	stk = stk[['Date','Adj. Open','Adj. High','Adj. Low','Adj. Close', 'Volume']]

	stk['Date'] = pandas.to_datetime(stk['Date'])    
	stk['Date'] = (stk['Date'] - stk['Date'].min())  / np.timedelta64(1,'D')


	#The column that needs to be forcasted using linear regression
	forecast_col = 'Adj. Close'
	
	#take care of NA's
	stk.fillna(-999999, inplace = True)
	stk['label'] = stk[forecast_col]


	#IN CASE THE INPUT IS TO BE TAKEN IN FROM THE COMMAND PROMPT UNCOMMENT THE LINES BELOW

	#takes in input from the user
	#x = list(map(int, input("Enter x: \n").split()))
	#y = list(map(int, input("Enter y: \n").split()))

	#convert to an numpy array with datatype as 64 bit float.
	#x = np.array(x, dtype = np.float64)
	#y = np.array(y, dtype = np.float64)

	stk.dropna(inplace = True)

	x = np.array(stk['Date'])
	y = np.array(stk['label'])

	#Always in the order: first slope, then intercept
	slope = simpl_linear_regression.best_fit(x, y) #find slope
	intercept = simpl_linear_regression.y_intercept(x, y) #find the intercept

	ip = list(map(int, input("Enter x to predict y: \n").split()))

	line = simpl_linear_regression.predict(ip) #predict based on model
	
	reg = [(slope*param) + intercept for param in x]

	print("Predicted value(s) after linear regression :", line)

	r_sqrd = simpl_linear_regression.cod(y, reg)
	print("R^2 Value: " ,r_sqrd)
	
	plt.scatter(x, y)
	plt.scatter(ip, line, color = "red")
	plt.plot(x, reg)
	plt.show()

if __name__ == "__main__":
	main()